MULTIVARIATE STATISTICAL ANALYSIS OF THE QUALITY OF BIOMETRIC DATA ON EXTREMELY SMALL
SAMPLES USING THE CRITERIA OF THE GEOMETRIC MEAN CALCULATED
FOR THE ANALYZED PROBABILITY FUNCTIONS

Ivanov Aleksandr Ivanovich, doctor of technical sciences, associate professor, head of biometric and neuronal nets technology laboratory, Penza Scientific Research Electrotechnical Institute (9 Sovetskaya street, Penza, Russia), ivan@pniei.penza.ru
Perfilov Konstantin Aleksandrovich, postgraduate student, Penza State University (40 Krasnaya street, Penza, Russia), ivan@pniei.penza.ru
Malygina Elena Aleksandrovna, candidate of technical sciences, research assistant, interdisciplinary laboratory testing of biometric devices and technologies, Penza State University (40 Krasnaya street, Penza, Russia), ivan@pniei.penza.ru

519.24; 519.7; 57.017

Background. The aim is to assess the capacity of the two variants of the statistical criteri of the geometric mean of the empirical and theoretical probability functions.
Materials and methods. It is proposed to use simulation tools and numerically to estimate power at comparable criteria equally errors of the first and second kind. Applying a logarithmic scale comparative assessment of capacities, which, depending on the number of power compared experiences in the training set are close to linear.
Results and conclusions. It is shown that the statistical criterion proposed by the geometric mean of the previously occupies an intermediate position between its two counterparts. The greatest power of the suppression of the quantization noise has a test based as the square root of the geometric mean of comparable probability functions. Considered criteria in their multidimensional embodiment capable of operating at extremely small samples of biometric data from 11 to 21, an example of a biometric image.

statistical criterion geometric mean of comparable probability functions, logarithmic scale power of statistical tests, multidimensional processing of biometric data, the quantization noise generated by the suppression of a small volume of the test sample.